Stacked as GeForce RTX 5090 specs are, it’s easy to forget that it doesn’t actually feature a full-fat GB202 GPU at its core. This potentially leaves the door open for a more-powerful card still, be it a GeForce RTX 5090 Ti or the long-awaited return of a ‘Titan’ SKU. It appears that such a potential SKU has slipped outside of Nvidia’s labs.
On the Chiphell forums, user omonob shares a picture of a regular-size PCB filled with an Nvidia GB202-200-A1 GPU, 16 GDDR7 memory chips, and two 12V-2×6 power headers. The size of the board is important to note as GeForce RTX 5090, Founders Edition or otherwise, is much smaller in design. Though the majority of the image is a blur, it’s still clear enough to distinguish the massive power delivery system.
According to omonob, this 24,576 CUDA core GPU and its accompanying 32GB of VRAM, operating at 32Gb/s for a total bandwidth of 2TB/s, carry an 800W TDP. Clock speeds are higher too, starting at 2.10GHz base and rising to 2.51GHz boost. For context, GeForce RTX 5090 aims for a more frugal 575W and clocks as 2.01-2.41GHz.
It’s impossible to conclude how much faster this prototype would be relative to GeForce RTX 5090 but the former does boast a 13% advantage in terms of CUDA core count. omonob indicates that there are no compatible drivers for this card despite its manufacturing date of July 2024. For now, this goliath will sadly continue to slumber.
This could be an early design from Nvidia to find the sweet spot for its consumer GPUs, leaving the fully unlocked chips for its higher-margin professional products. While we would love to see a consumer version of this card, chances are it will probably end up like the RTX 40 prototype. Moreover, considering GeForce RTX 5090’s $1,999 MSRP, a fully unlocked chip with faster VRAM would easily ask for $2,500 if not more. Let’s not give anyone any ideas.
